Selfish Overlay Network Formation Georgios Smaragdakis 1 1 - PowerPoint PPT Presentation

Selfish Overlay Network Formation Georgios Smaragdakis 1 1

Deutsche Telekom Laboratories. T-Labs, An-Institute of Technische Universität Berlin T-Labs, Ben-Gurion University T-Labs US, Stanford University 2 2

Strategic Research concentrates on long- term technology and applied research. Service-centric Quality and Security in Intelligent Networks Usability Lab Telecommunications Networking  Network  Audio  Microkernel  Definition in Measurement Technology Security 2010 and Security  Image and  Vehicular  Routing Vision Security  Wireless Computing  Wireless  Mobile and Networks Security  Virtualization Physical  Server Security Interaction  Peer to Peer  Data Security  Quality  Content and  Speech Cryptography Distribution Technology Networks  Usability  Functional groups Networking / Security / Usability 3 3

Innovation Development and Strategic Research work side by side, to jointly achieve goals. Strategic Research Innovation Development  Strategic Research concentrates on  Innovation Development develops the long-term technology research innovative solutions, as a basis for and applied research. the commercial use by the Group‘s business areas.  Strategic Research creates the foundation for the development of innovative solutions in Innovation Development.  Results  Results  Publications Market studies   Patents Acceptance tests   Demonstrations Business models  Prototypes  4 4

The success of Telekom Laboratories is measured at the transfer to the Group’s business areas or to spin-offs. 5 5

Telekom Laboratories cooperate according to the Open Innovation model with selected Rheinische Friedrich- Wilhelms-Universität Norwegian University of research institutes. Bonn Science and Technology Imperial College Technische Universität London Berlin École Nationale Princeton d’Ingénieurs de University Fraunhofer-Institut für Brest Nachrichtentechnik Boston Univeridad Heinrich-Hertz-Institut University Carlos III de Fraunhofer-Institut für University of Madrid Offene Illinois Technische Kommunikationssystem Universität e Darmstadt Ben-Gurion University Stanford University Universite Catholique Ludwig-Maximilian- de Louvain UC Universität München Berkeley/ICSI École Polytechnique Technische Universität Fédérale de Lausanne München Universität St. Gallen 6 6

Selfish Overlay Network Formation Georgios Smaragdakis Joint work with Nikolaos Laoutaris, Azer Bestavros, John Byers, Pietro Michiardi, Mema Roussopoulos and Vassilis Lekakis 7 1

O ver l ays overlay O 2 plane process O 1 O 3 overlay node physical plane R 3 R 2 R 1 router, AS R 4 8 2

O ver l ay Econom i cs & Nei ghbor Sel ect i on Market: Investment: Flat Resource Allocation i nt e r ne t t r a ns i t I SP t r a ns i t I SP overlay links $ $$ $$ $$ a c c e s s I SP overlay node a c c e s s I SP a c c e s s I SP 9 3

Connect i vi t y M anagem ent  Full mesh architectures for reliability (e.g. RON)  Myopic heuristics random or proximity based neighbor selection  Tree forest or mesh construction to optimize multicast (e.g. Bullet, Splitstream)  Optimization for network delay (e.g. Detour, QRON)  Opportunistic choke/unchoke (e.g. BitTorrent)  Distributed hashing tables (e.g. Chord, Pastry, Tapestry) 10 4

Chal l enges Op p o r t uni t i e s  Network Heterogeneity: pair wise delay or available bandwidth, storage, cpu cycles, budget…  Load Variability: diurnal variation of traffic, dynamic routing or pricing, node churn…  Diversity of users: different prospective, conflicting objectives 11 5

St r at egi c Resour ce Al l ocat i on Tr a ns a c t i o n o n I NFOCOM ’ 0 7 Ne t wo r k i ng I m pl i cat i ons t o Sel f i sh O ver l ay Pr ot ocol Desi gn Net w or k Cr eat i on EG O I ST Appl i cat i on t o sw ar m i ng syst em s I nf o c o m 2 0 0 8 , TPDS Co NEXT 2 0 0 8 12 6

St r at egi c Resour ce Al l ocat i on I m pl i cat i ons t o Sel f i sh O ver l ay Pr ot ocol Desi gn Net w or k Cr eat i on EG O I ST Appl i cat i on t o sw ar m i ng syst em s 13 7

Net w or k Cr eat i on Local Connection Game <V,{s i },{C i }> [Fabrikant et al,PODC’03]  V: set of n players (nodes)  {s i }: strategies available to v i (wirings)  {C i }: set of utilities for v i (cost)  Outcome: S is the global wiring a a ∑ = α ⋅ + c S s d v v ( ) | | ( , ) m i n i i S i j ∈ v j V − i 14 8

O ver l ay Net w or k Cr eat i on Towards a Real c model for Overlay Networks: i st i  Directed Edges  Bounded out- and in-degree  Non-uniform preference vectors  Realistic models of physical distance Towards a Ri G am e , easi e via a network cher l y r eal i zabl protocol. 15 9

Sel f i sh Nei ghbor Sel ect i on ( SNS) v i : Choose k neighbors ∑ = ⋅ C S p d v v ( ) ( , ) m i n i ij S i j ∈ w v j V − i ∈ S o v e r a l l s i i u v i G - =( V - , S - ) i i i Se t o f r e s i d ua l no d e s Se t o f r e s i d ua l wi r i ng v i ’ s r esi dual net w or k 16 10

SNS & k- m edi an Uniform link weights, and uniform preference  k-median on asymmetric distances 17 11

k- m edi an k- m edi an: Find a subset I of F and a function σ:C  I, to: min ( Σ i,j s j c ij ) such that |I| ≤ k F: set of C: set of clients, facilities c ij : cost connecting client j  facility I s j : demand of node 18 12

k- m edi an 19 13

SNS & k- m edi an Uniform link weights, and uniform preference  k-median on asymmetric distances Si nce t he w i r i ng w w Non-uniform link weights, and uniform cost i s t he sam e preference v i  ILP formulation u u ∑ = ⋅ w , u can be C ( S ) p d ( v , v ) m i n i ij S i j obt ai ned f r om ∈ v j V − i k- m edi an on r ever sed di st ances 20 14

Local Sear ch ( LS) v i : choose k neighbors ∑ = ⋅ C S p d v v ( ) ( , ) m i n i ij S i j ∈ w v j V − i ∈ S o v e r a l l s i i u v i [Arya et al,STOC’01] G - =( V - , S - ) i i i Se t o f r e s i d ua l no d e s Se t o f r e s i d ua l wi r i ng v i ’ s r esi dual net w or k 21 15

SNS : t he G am e Game <V,{s i },{C i }>  V : set of n players (nodes)  {s i }: strategies available to v i (wirings), choose k out of n to connect  {C i }: set of costs for v i ∑ = ⋅ min C S p d v v ( ) ( , ) i ij S i j ∈ v j V − i Best response of a node: node’s optimal wiring Outcome: S, the global wiring  A stable wiring is a pure Nash equilibium  Using iterative best response  Fundamentally different from selfish routing 22 16

SNS : Equi l i br i a Uniform Preference Skewness of preference n=15 k=2 k=3 k=8 I n- degr ees ar e hi ghl y skew ed k=11 even under uni f or m pr ef er ence !  Qua l i t y - b a s e d “ p r e f e r e nt i a l a t t a c h me nt ” 23 k (Link density) 17

SNS : Ef f i ci ency Performance of ILP & LS is close to Utopian! Skewness of Skewness of Link density preference Link density preference Theoretical results showed in the worst case the social cost can be bad 24 [Laoutaris, Poplawsi, Rajaraman, Sundaram, Teng,PODC’08] 18

St r at egi c Resour ce Al l ocat i on I m pl i cat i ons t o Sel f i sh O ver l ay Pr ot ocol Desi gn Net w or k Cr eat i on EG O I ST Appl i cat i on t o sw ar m i ng syst em s 25 19

SNS : Tr ace- Dr i ven Eval uat i on How we assign the distance:  Synthetically using BRITE  Empirically from PlanetLab  Empirically from AS-level maps [Routeviews] Neighbor Selection Strategies:  k-Random heuristic  k-Closest heuristic  k-Regular heuristic  k-Best Response Control parameter:  Bound on out-degree k (link density) 26 20

SNS vs. Heur i st i cs: Soci al Cost Macroscopic view: Focusing on the social welfare (k=2) k-Random/BR k-Closest/BR k-Regular/BR BRITE 1.44 1.53 3.61 PlanetLab 2.23 1.48 3.84 AS 2.04 1.90 4.78 The network is better off with selfish nodes! 27 21

Connect i ng on a k- Random gr aph PlanetLab ( n =50 ) AS-Level ( n =50 ) BRITE ( n =50 ) 0 2 3 5 11 22 0 2 3 5 11 22 0 2 3 5 11 22 k k k If your neighbors are naïve, it pays to be selfish! 28 22

Connect i ng on a k- Cl osest gr aph PlanetLab ( n =50 ) AS-Level ( n =50 ) BRITE ( n =50 ) 0 2 3 5 11 22 0 2 3 5 11 22 0 2 3 5 11 22 k k k If your neighbors are greedy, it pays to be selfish! 29 23